Self-supervised Learning for Hyperspectral Images of Trees
This work addresses a domain-specific problem in precision agriculture for analyzing tree vegetation from aerial hyperspectral images, but it is incremental as it builds on existing self-supervised learning methods.
The paper tackled the challenge of analyzing hyperspectral images with limited labels by using self-supervised learning to create neural network embeddings for trees, resulting in improved performance in downstream tasks compared to using direct hyperspectral properties.
Aerial remote sensing using multispectral and RGB imagers has provided a critical impetus to precision agriculture. Analysis of the hyperspectral images with limited or no labels is challenging. This paper focuses on self-supervised learning to create neural network embeddings reflecting vegetation properties of trees from aerial hyperspectral images of crop fields. Experimental results demonstrate that a constructed tree representation, using a vegetation property-related embedding space, performs better in downstream machine learning tasks compared to the direct use of hyperspectral vegetation properties as tree representations.